Giving a fleet of AI agents one shared memory — when each agent runs a different model

Most agent frameworks give each agent its own context window and call it memory. That works right...